ewoooc/routes/ai_routes.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
AI 推薦路由模組
提供時事熱點商品推薦與文案生成功能
支援 Ollama 和 Gemini 雙 AI 提供者
"""

from flask import Blueprint, render_template, request, jsonify, session
from auth import login_required
from services.ollama_service import OllamaService
from services.gemini_service import GeminiService, AVAILABLE_GEMINI_MODELS, GEMINI_PRICING
from services.ai_provider import AIProviderService, ai_provider_service, get_ai_status, set_ai_provider
from services.trend_crawler import TrendCrawler
from services.ai_history_service import AIHistoryService, AITemplateService
from database.manager import DatabaseManager
from database.ai_models import AIUsageTracking
import pandas as pd
import logging
import json
import os
from datetime import datetime, date, timedelta

logger = logging.getLogger(__name__)

# 分類 JSON 路徑
CATEGORIES_JSON_PATH = os.path.join(os.path.dirname(os.path.dirname(__file__)), 'data', 'categories.json')

def load_product_categories():
    """從 JSON 檔案載入商品分類列表"""
    try:
        with open(CATEGORIES_JSON_PATH, 'r', encoding='utf-8') as f:
            categories = json.load(f)
            return [c['name'] for c in categories]
    except (FileNotFoundError, json.JSONDecodeError):
        return []

ai_bp = Blueprint('ai', __name__)

# 服務實例
ollama_service = OllamaService()
gemini_service = GeminiService()
trend_crawler = TrendCrawler()
ai_history_service = AIHistoryService()
ai_template_service = AITemplateService()


@ai_bp.route('/ai_recommend')
@login_required
def ai_recommend():
    """AI 推薦頁面"""
    # 取得 AI 服務狀態
    ai_status = get_ai_status()

    # 載入商品分類（從商品看板設定）
    product_categories = load_product_categories()

    return render_template('ai_recommend.html',
                           ai_status=ai_status,
                           ollama_status=ai_status['ollama']['connected'],
                           gemini_status=ai_status['gemini']['connected'],
                           available_models=ai_status['ollama']['available_models'],
                           gemini_models=AVAILABLE_GEMINI_MODELS,
                           default_provider=ai_status['default_provider'],
                           product_categories=product_categories)


@ai_bp.route('/api/ai/status')
@login_required
def api_ai_status():
    """檢查 AI 服務狀態（雙提供者）"""
    force_refresh = request.args.get('refresh', 'false').lower() == 'true'
    status = get_ai_status(force_refresh=force_refresh)

    return jsonify({
        'success': True,
        'data': status
    })


@ai_bp.route('/api/ai/set_provider', methods=['POST'])
@login_required
def api_set_provider():
    """切換預設 AI 提供者"""
    try:
        data = request.get_json()
        provider = data.get('provider', 'ollama')

        if provider not in ('ollama', 'gemini'):
            return jsonify({'success': False, 'error': '無效的提供者，請使用 ollama 或 gemini'}), 400

        success = set_ai_provider(provider)
        if success:
            logger.info(f"AI 提供者已切換至: {provider}")
            return jsonify({
                'success': True,
                'message': f'已切換至 {provider.upper()}',
                'provider': provider
            })
        else:
            return jsonify({'success': False, 'error': '切換失敗'}), 500

    except Exception as e:
        logger.error(f"切換 AI 提供者失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/gemini_usage')
@login_required
def api_gemini_usage():
    """取得 Gemini API 使用量和費用統計"""
    try:
        days = request.args.get('days', 30, type=int)
        if days > 365:
            days = 365

        db = DatabaseManager()
        session = db.get_session()

        try:
            # 計算日期範圍
            end_date = date.today()
            start_date = end_date - timedelta(days=days)

            # 查詢使用量
            from sqlalchemy import func
            usage_query = session.query(
                AIUsageTracking.model_name,
                func.sum(AIUsageTracking.input_tokens).label('total_input_tokens'),
                func.sum(AIUsageTracking.output_tokens).label('total_output_tokens'),
                func.sum(AIUsageTracking.total_tokens).label('total_tokens'),
                func.sum(AIUsageTracking.total_cost).label('total_cost'),
                func.count(AIUsageTracking.id).label('request_count')
            ).filter(
                AIUsageTracking.provider == 'gemini',
                AIUsageTracking.request_date >= start_date,
                AIUsageTracking.request_date <= end_date
            ).group_by(AIUsageTracking.model_name).all()

            # 計算總計
            total_cost = 0.0
            total_tokens = 0
            total_requests = 0
            by_model = []

            for row in usage_query:
                model_data = {
                    'model': row.model_name,
                    'input_tokens': row.total_input_tokens or 0,
                    'output_tokens': row.total_output_tokens or 0,
                    'total_tokens': row.total_tokens or 0,
                    'total_cost': float(row.total_cost or 0),
                    'request_count': row.request_count or 0
                }
                by_model.append(model_data)
                total_cost += model_data['total_cost']
                total_tokens += model_data['total_tokens']
                total_requests += model_data['request_count']

            # 每日使用趨勢
            daily_query = session.query(
                AIUsageTracking.request_date,
                func.sum(AIUsageTracking.total_cost).label('daily_cost'),
                func.count(AIUsageTracking.id).label('daily_requests')
            ).filter(
                AIUsageTracking.provider == 'gemini',
                AIUsageTracking.request_date >= start_date,
                AIUsageTracking.request_date <= end_date
            ).group_by(AIUsageTracking.request_date).order_by(AIUsageTracking.request_date).all()

            daily_trend = [
                {
                    'date': row.request_date.isoformat(),
                    'cost': float(row.daily_cost or 0),
                    'requests': row.daily_requests or 0
                }
                for row in daily_query
            ]

            return jsonify({
                'success': True,
                'data': {
                    'period': {
                        'start': start_date.isoformat(),
                        'end': end_date.isoformat(),
                        'days': days
                    },
                    'summary': {
                        'total_cost_usd': round(total_cost, 6),
                        'total_tokens': total_tokens,
                        'total_requests': total_requests
                    },
                    'by_model': by_model,
                    'daily_trend': daily_trend,
                    'pricing': GEMINI_PRICING
                }
            })

        finally:
            session.close()

    except Exception as e:
        logger.error(f"取得 Gemini 使用量失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/trends')
@login_required
def api_get_trends():
    """獲取趨勢資料"""
    try:
        categories = request.args.getlist('categories') or ['時尚美妝', '生活居家', '健康保健']
        location = request.args.get('location', '臺北市')
        time_range = request.args.get('time_range', 'week')  # day, week, month
        include_social = request.args.get('include_social', 'true').lower() == 'true'

        # 驗證時間範圍參數
        if time_range not in ['day', 'week', 'month']:
            time_range = 'week'

        trend_data = trend_crawler.get_all_trends(
            categories=categories,
            weather_location=location,
            time_range=time_range,
            include_social=include_social
        )

        # 轉換為可序列化的格式
        result = {
            'timestamp': trend_data.timestamp.isoformat(),
            'time_range': time_range,
            'news': [
                {
                    'title': n.title,
                    'link': n.link,
                    'source': n.source,
                    'category': n.category,
                    'published': n.published.isoformat()
                }
                for n in trend_data.news_items[:30]  # 最多 30 則
            ],
            'youtube': [
                {
                    'title': v.title,
                    'video_id': v.video_id,
                    'url': v.url,
                    'channel': v.channel_title,
                    'category': v.category,
                    'thumbnail': v.thumbnail_url,
                    'published': v.published.isoformat()
                }
                for v in trend_data.youtube_videos[:20]  # 最多 20 則
            ],
            'social': [
                {
                    'title': p.title,
                    'link': p.link,
                    'source': p.source,
                    'board': p.board,
                    'category': p.category,
                    'likes': p.likes,
                    'comments': p.comments,
                    'published': p.published.isoformat()
                }
                for p in trend_data.social_posts[:30]  # 最多 30 則
            ],
            'weather': None,
            'keywords': trend_data.keywords[:15],
            'category_trends': {
                cat: titles[:5]
                for cat, titles in trend_data.category_trends.items()
            }
        }

        if trend_data.weather:
            result['weather'] = {
                'location': trend_data.weather.location,
                'date': trend_data.weather.date,
                'description': trend_data.weather.weather_description,
                'min_temp': trend_data.weather.min_temp,
                'max_temp': trend_data.weather.max_temp,
                'rain_probability': trend_data.weather.rain_probability,
                'humidity': trend_data.weather.humidity,
                'marketing_suggestions': trend_data.weather.marketing_suggestions
            }

        return jsonify({'success': True, 'data': result})

    except Exception as e:
        logger.error(f"獲取趨勢資料失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/weather')
@login_required
def api_get_weather():
    """獲取天氣資訊"""
    try:
        location = request.args.get('location', '臺北市')
        weather = trend_crawler.fetch_weather(location)

        if weather:
            return jsonify({
                'success': True,
                'data': {
                    'location': weather.location,
                    'date': weather.date,
                    'description': weather.weather_description,
                    'min_temp': weather.min_temp,
                    'max_temp': weather.max_temp,
                    'rain_probability': weather.rain_probability,
                    'humidity': weather.humidity,
                    'marketing_suggestions': weather.marketing_suggestions
                }
            })
        else:
            return jsonify({'success': False, 'error': '無法獲取天氣資訊'}), 500

    except Exception as e:
        logger.error(f"獲取天氣失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/generate_copy', methods=['POST'])
@login_required
def api_generate_copy():
    """生成銷售文案（支援 Ollama/Gemini 雙提供者）"""
    try:
        data = request.get_json()
        product_name = data.get('product_name', '')
        trend_keywords = data.get('trend_keywords', [])
        style = data.get('style', '吸睛')
        model = data.get('model', None)
        provider = data.get('provider', None)  # 'ollama' 或 'gemini'
        save_to_history = data.get('save_to_history', True)

        # 額外的上下文資訊
        upcoming_holidays = data.get('upcoming_holidays', [])
        bestseller_products = data.get('bestseller_products', [])

        if not product_name:
            return jsonify({'success': False, 'error': '請提供商品名稱'}), 400

        # 使用統一的 AI 提供者服務
        result = ai_provider_service.generate_sales_copy(
            product_name=product_name,
            provider=provider,
            model=model,
            trend_keywords=trend_keywords,
            style=style,
            upcoming_holidays=upcoming_holidays,
            bestseller_products=bestseller_products
        )

        if result.success:
            history_id = None
            user_id = session.get('user_id')

            # 儲存到歷史記錄
            if save_to_history:
                history_id = ai_history_service.save_generation(
                    generation_type='copy',
                    output_content=result.content,
                    product_name=product_name,
                    input_keywords=trend_keywords,
                    input_style=style,
                    model_name=result.model,
                    generation_duration=result.total_duration,
                    created_by=user_id,
                    ai_provider=result.provider,
                    input_tokens=result.input_tokens,
                    output_tokens=result.output_tokens
                )

            # 如果是 Gemini，記錄使用量
            if result.provider == 'gemini' and result.total_tokens > 0:
                _save_gemini_usage(
                    model_name=result.model,
                    usage_type='copy',
                    input_tokens=result.input_tokens,
                    output_tokens=result.output_tokens,
                    total_cost=result.total_cost,
                    duration=result.total_duration,
                    user_id=user_id,
                    history_id=history_id
                )

            return jsonify({
                'success': True,
                'data': {
                    'copy': result.content,
                    'model': result.model,
                    'provider': result.provider,
                    'duration': round(result.total_duration, 2) if result.total_duration else None,
                    'history_id': history_id,
                    'tokens': {
                        'input': result.input_tokens,
                        'output': result.output_tokens,
                        'total': result.total_tokens
                    },
                    'cost': {
                        'input': result.input_cost,
                        'output': result.output_cost,
                        'total': result.total_cost
                    } if result.provider == 'gemini' else None
                }
            })
        else:
            return jsonify({'success': False, 'error': result.error}), 500

    except Exception as e:
        logger.error(f"生成文案失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


def _save_gemini_usage(model_name: str, usage_type: str, input_tokens: int,
                       output_tokens: int, total_cost: float, duration: float,
                       user_id: int = None, history_id: int = None):
    """儲存 Gemini API 使用記錄"""
    try:
        db = DatabaseManager()
        db_session = db.get_session()

        try:
            usage = AIUsageTracking(
                provider='gemini',
                model_name=model_name,
                usage_type=usage_type,
                input_tokens=input_tokens,
                output_tokens=output_tokens,
                total_tokens=input_tokens + output_tokens,
                input_cost=GeminiService.calculate_cost(model_name, input_tokens, 0)['input_cost'],
                output_cost=GeminiService.calculate_cost(model_name, 0, output_tokens)['output_cost'],
                total_cost=total_cost,
                duration=duration,
                request_date=date.today(),
                created_by=user_id,
                history_id=history_id
            )
            db_session.add(usage)
            db_session.commit()
            logger.info(f"Gemini 使用記錄已儲存: {model_name}, ${total_cost:.6f}")
        finally:
            db_session.close()

    except Exception as e:
        logger.error(f"儲存 Gemini 使用記錄失敗: {e}")


@ai_bp.route('/api/ai/recommend_products', methods=['POST'])
@login_required
def api_recommend_products():
    """根據趨勢推薦商品"""
    try:
        data = request.get_json()
        trend_topic = data.get('trend_topic', '')
        trend_description = data.get('trend_description', '')

        if not trend_topic:
            return jsonify({'success': False, 'error': '請提供趨勢話題'}), 400

        # 從資料庫獲取商品列表
        db = DatabaseManager()
        engine = db.engine

        # 查詢商品
        try:
            df = pd.read_sql("""
                SELECT DISTINCT
                    商品名稱 as name,
                    商品分類L1 as category_l1,
                    商品分類L2 as category_l2,
                    品牌 as brand
                FROM products
                WHERE 商品名稱 IS NOT NULL
                LIMIT 100
            """, engine)

            products = df.to_dict('records')
        except Exception as e:
            # 如果 products 表不存在，嘗試從 daily_sales_snapshot 取得
            logger.warning(f"products 表查詢失敗: {e}, 嘗試 daily_sales_snapshot")
            try:
                df = pd.read_sql("""
                    SELECT DISTINCT
                        商品名稱 as name,
                        商品分類L1 as category_l1,
                        商品分類L2 as category_l2,
                        品牌 as brand
                    FROM daily_sales_snapshot
                    WHERE 商品名稱 IS NOT NULL
                    LIMIT 100
                """, engine)
                products = df.to_dict('records')
            except Exception as e2:
                logger.error(f"查詢商品失敗: {e2}")
                products = []

        if not products:
            return jsonify({'success': False, 'error': '沒有可用的商品資料'}), 400

        # 使用 LLM 進行匹配
        result = ollama_service.match_products_to_trend(
            trend_topic=trend_topic,
            trend_description=trend_description,
            products=[
                {
                    'name': p.get('name', ''),
                    'category': f"{p.get('category_l1', '')} > {p.get('category_l2', '')}".strip(' > '),
                    'brand': p.get('brand', '')
                }
                for p in products
            ]
        )

        if result.success:
            # 嘗試解析 JSON 結果
            try:
                # 提取 JSON 部分
                content = result.content
                json_match = content[content.find('{'):content.rfind('}') + 1]
                recommendations = json.loads(json_match)

                return jsonify({
                    'success': True,
                    'data': {
                        'recommendations': recommendations.get('recommendations', []),
                        'model': result.model,
                        'duration': round(result.total_duration, 2) if result.total_duration else None
                    }
                })
            except json.JSONDecodeError:
                # 如果無法解析 JSON，返回原始內容
                return jsonify({
                    'success': True,
                    'data': {
                        'raw_response': result.content,
                        'model': result.model
                    }
                })
        else:
            return jsonify({'success': False, 'error': result.error}), 500

    except Exception as e:
        logger.error(f"推薦商品失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/analyze_weather_products', methods=['POST'])
@login_required
def api_analyze_weather_products():
    """根據天氣分析適合推薦的商品類型"""
    try:
        data = request.get_json()
        location = data.get('location', '臺北市')

        # 獲取天氣
        weather = trend_crawler.fetch_weather(location)

        if not weather:
            return jsonify({'success': False, 'error': '無法獲取天氣資訊'}), 500

        # 建立分析提示
        weather_context = f"""
今日天氣資訊（{weather.location}）：
- 天氣狀況：{weather.weather_description}
- 溫度：{weather.min_temp}°C ~ {weather.max_temp}°C
- 降雨機率：{weather.rain_probability}%
- 濕度：{weather.humidity}%
"""

        system_prompt = """你是一位電商行銷專家，擅長根據天氣狀況推薦適合的商品類型。
請根據天氣資訊，分析今天適合推薦哪些類型的商品，並說明原因。"""

        prompt = f"""{weather_context}

請分析今天的天氣狀況，推薦適合銷售的商品類型。
請用以下 JSON 格式回覆：
{{
  "weather_summary": "天氣簡述",
  "recommended_categories": [
    {{"category": "商品類型", "reason": "推薦原因", "keywords": ["關鍵字1", "關鍵字2"]}}
  ],
  "marketing_tips": ["行銷建議1", "行銷建議2"]
}}"""

        result = ollama_service.generate(prompt, system_prompt=system_prompt, temperature=0.5)

        if result.success:
            try:
                content = result.content
                json_match = content[content.find('{'):content.rfind('}') + 1]
                analysis = json.loads(json_match)

                return jsonify({
                    'success': True,
                    'data': {
                        'weather': {
                            'location': weather.location,
                            'description': weather.weather_description,
                            'temp_range': f"{weather.min_temp}°C ~ {weather.max_temp}°C",
                            'rain_probability': weather.rain_probability,
                            'humidity': weather.humidity,
                            'builtin_suggestions': weather.marketing_suggestions
                        },
                        'analysis': analysis,
                        'model': result.model
                    }
                })
            except json.JSONDecodeError:
                return jsonify({
                    'success': True,
                    'data': {
                        'weather': {
                            'location': weather.location,
                            'description': weather.weather_description,
                            'temp_range': f"{weather.min_temp}°C ~ {weather.max_temp}°C",
                            'rain_probability': weather.rain_probability,
                            'humidity': weather.humidity,
                            'builtin_suggestions': weather.marketing_suggestions
                        },
                        'raw_response': result.content
                    }
                })
        else:
            return jsonify({'success': False, 'error': result.error}), 500

    except Exception as e:
        logger.error(f"天氣商品分析失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/batch_generate_copy', methods=['POST'])
@login_required
def api_batch_generate_copy():
    """批次生成多個商品的文案"""
    try:
        data = request.get_json()
        products = data.get('products', [])  # [{"name": "...", "keywords": [...]}]
        style = data.get('style', '吸睛')

        if not products:
            return jsonify({'success': False, 'error': '請提供商品列表'}), 400

        if len(products) > 10:
            return jsonify({'success': False, 'error': '一次最多處理 10 個商品'}), 400

        results = []
        for product in products:
            result = ollama_service.generate_sales_copy(
                product_name=product.get('name', ''),
                trend_keywords=product.get('keywords', []),
                style=style
            )

            results.append({
                'product_name': product.get('name', ''),
                'success': result.success,
                'copy': result.content if result.success else None,
                'error': result.error if not result.success else None
            })

        success_count = sum(1 for r in results if r['success'])

        return jsonify({
            'success': True,
            'data': {
                'results': results,
                'total': len(results),
                'success_count': success_count
            }
        })

    except Exception as e:
        logger.error(f"批次生成文案失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


# ===== AI 歷史記錄 CRUD API =====

@ai_bp.route('/ai_history')
@login_required
def ai_history_page():
    """AI 歷史記錄頁面"""
    return render_template('ai_history.html')


@ai_bp.route('/api/ai/history')
@login_required
def api_get_history():
    """獲取 AI 生成歷史列表"""
    try:
        page = request.args.get('page', 1, type=int)
        per_page = request.args.get('per_page', 20, type=int)
        generation_type = request.args.get('type')
        search = request.args.get('search')
        is_favorite = request.args.get('favorite')
        start_date = request.args.get('start_date')
        end_date = request.args.get('end_date')

        # 處理布林值
        if is_favorite is not None:
            is_favorite = is_favorite.lower() == 'true'

        # 處理日期
        from datetime import datetime
        if start_date:
            start_date = datetime.fromisoformat(start_date)
        if end_date:
            end_date = datetime.fromisoformat(end_date)

        result = ai_history_service.get_history_list(
            page=page,
            per_page=per_page,
            generation_type=generation_type,
            search=search,
            is_favorite=is_favorite,
            start_date=start_date,
            end_date=end_date
        )

        return jsonify({'success': True, 'data': result})

    except Exception as e:
        logger.error(f"獲取 AI 歷史記錄失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/history/<int:history_id>')
@login_required
def api_get_history_detail(history_id):
    """獲取單筆歷史記錄"""
    try:
        history = ai_history_service.get_by_id(history_id)
        if history:
            return jsonify({'success': True, 'data': history})
        else:
            return jsonify({'success': False, 'error': '記錄不存在'}), 404
    except Exception as e:
        logger.error(f"獲取 AI 歷史記錄詳情失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/history/<int:history_id>', methods=['PUT'])
@login_required
def api_update_history(history_id):
    """更新歷史記錄"""
    try:
        data = request.get_json()
        success = ai_history_service.update(
            history_id=history_id,
            output_content=data.get('output_content'),
            rating=data.get('rating'),
            is_favorite=data.get('is_favorite'),
            is_used=data.get('is_used'),
            notes=data.get('notes')
        )
        if success:
            return jsonify({'success': True, 'message': '更新成功'})
        else:
            return jsonify({'success': False, 'error': '記錄不存在或更新失敗'}), 404
    except Exception as e:
        logger.error(f"更新 AI 歷史記錄失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/history/<int:history_id>', methods=['DELETE'])
@login_required
def api_delete_history(history_id):
    """刪除歷史記錄"""
    try:
        success = ai_history_service.delete(history_id)
        if success:
            return jsonify({'success': True, 'message': '刪除成功'})
        else:
            return jsonify({'success': False, 'error': '記錄不存在或刪除失敗'}), 404
    except Exception as e:
        logger.error(f"刪除 AI 歷史記錄失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/history/batch', methods=['DELETE'])
@login_required
def api_batch_delete_history():
    """批次刪除歷史記錄"""
    try:
        data = request.get_json()
        ids = data.get('ids', [])
        if not ids:
            return jsonify({'success': False, 'error': '請提供要刪除的 ID 列表'}), 400

        result = ai_history_service.batch_delete(ids)
        return jsonify({
            'success': True,
            'data': result
        })
    except Exception as e:
        logger.error(f"批次刪除 AI 歷史記錄失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/history/<int:history_id>/favorite', methods=['POST'])
@login_required
def api_toggle_favorite(history_id):
    """切換收藏狀態"""
    try:
        new_status = ai_history_service.toggle_favorite(history_id)
        if new_status is not None:
            return jsonify({
                'success': True,
                'data': {'is_favorite': new_status}
            })
        else:
            return jsonify({'success': False, 'error': '記錄不存在'}), 404
    except Exception as e:
        logger.error(f"切換收藏狀態失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/statistics')
@login_required
def api_get_statistics():
    """獲取 AI 統計資料"""
    try:
        days = request.args.get('days', 30, type=int)
        stats = ai_history_service.get_statistics(days=days)
        return jsonify({'success': True, 'data': stats})
    except Exception as e:
        logger.error(f"獲取 AI 統計資料失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


# ===== AI 模板 API =====

@ai_bp.route('/api/ai/templates')
@login_required
def api_get_templates():
    """獲取 AI 提示模板列表"""
    try:
        template_type = request.args.get('type')
        templates = ai_template_service.get_all_templates(template_type=template_type)
        return jsonify({'success': True, 'data': templates})
    except Exception as e:
        logger.error(f"獲取 AI 模板列表失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/templates/<int:template_id>')
@login_required
def api_get_template(template_id):
    """獲取單個模板"""
    try:
        template = ai_template_service.get_by_id(template_id)
        if template:
            return jsonify({'success': True, 'data': template})
        else:
            return jsonify({'success': False, 'error': '模板不存在'}), 404
    except Exception as e:
        logger.error(f"獲取 AI 模板失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


# ===== 熱銷商品 API =====

@ai_bp.route('/api/ai/bestsellers')
@login_required
def api_get_bestsellers():
    """取得熱銷商品（支援 PChome 和 MOMO）"""
    try:
        category = request.args.get('category', '面膜')
        limit = request.args.get('limit', 5, type=int)
        platform = request.args.get('platform', 'pchome').lower()  # pchome 或 momo

        if limit > 10:
            limit = 10  # 限制最多 10 個

        if platform == 'momo':
            from services.momo_crawler import get_momo_bestsellers
            success, message, products = get_momo_bestsellers(category, limit=limit)
            source = 'MOMO購物網'
        else:
            from services.pchome_crawler import get_pchome_bestsellers
            success, message, products = get_pchome_bestsellers(category, limit=limit)
            source = 'PChome 24h'

        if success:
            return jsonify({
                'success': True,
                'data': {
                    'category': category,
                    'products': products,
                    'source': source,
                    'platform': platform
                }
            })
        else:
            return jsonify({'success': False, 'error': message}), 500

    except Exception as e:
        logger.error(f"取得熱銷商品失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/sales_suggestion', methods=['POST'])
@login_required
def api_generate_sales_suggestion():
    """生成商品銷售策略建議"""
    try:
        data = request.get_json()
        product_name = data.get('product_name', '')
        trend_keywords = data.get('trend_keywords', [])
        upcoming_holidays = data.get('upcoming_holidays', [])
        bestseller_products = data.get('bestseller_products', [])
        model = data.get('model', None)

        if not product_name:
            return jsonify({'success': False, 'error': '請提供商品名稱'}), 400

        # 建立上下文
        context_parts = []

        if trend_keywords:
            context_parts.append(f"熱門趨勢關鍵字：{', '.join(trend_keywords)}")

        if upcoming_holidays:
            holidays_text = []
            for h in upcoming_holidays[:3]:
                name = h.get('name', '')
                days = h.get('days_until', 0)
                if days == 0:
                    holidays_text.append(f"{name}（今天）")
                elif days == 1:
                    holidays_text.append(f"{name}（明天）")
                else:
                    holidays_text.append(f"{name}（{days}天後）")
            if holidays_text:
                context_parts.append(f"即將到來的節日：{', '.join(holidays_text)}")

        if bestseller_products:
            products_text = [f"{p.get('name', '')}（${p.get('price', '')}）" for p in bestseller_products[:5]]
            if products_text:
                context_parts.append(f"市場競品參考：{', '.join(products_text)}")

        context = '\n'.join(context_parts) if context_parts else '無額外市場資訊'

        system_prompt = """你是一位資深電商銷售策略顧問，專精於台灣市場。
你的建議特點：
- 精準、可執行的銷售策略
- 考慮節日、季節、市場趨勢
- 針對競品進行差異化定位
- 提供具體的促銷方案建議"""

        prompt = f"""請為以下商品提供銷售策略建議：

商品名稱：{product_name}

市場資訊：
{context}

請提供：
1. 【定位建議】這個商品應該如何定位？（1-2句）
2. 【目標客群】主要目標客群是誰？
3. 【促銷策略】針對近期節日或趨勢的促銷建議（2-3點）
4. 【差異化賣點】相較競品的差異化賣點建議

請簡潔回答，每點不超過 30 字。"""

        # 如果指定了模型，暫時切換
        original_model = ollama_service.model
        if model:
            ollama_service.model = model

        result = ollama_service.generate(prompt, system_prompt=system_prompt, temperature=0.6)

        # 恢復原始模型
        if model:
            ollama_service.model = original_model

        if result.success:
            return jsonify({
                'success': True,
                'data': {
                    'suggestion': result.content,
                    'model': result.model,
                    'duration': round(result.total_duration, 2) if result.total_duration else None
                }
            })
        else:
            return jsonify({'success': False, 'error': result.error}), 500

    except Exception as e:
        logger.error(f"生成銷售建議失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/cosme_rankings')
@login_required
def api_get_cosme_rankings():
    """取得 COSME 排行榜"""
    try:
        category = request.args.get('category', 'mask')
        limit = request.args.get('limit', 10, type=int)

        if limit > 20:
            limit = 20

        from services.cosme_crawler import get_cosme_rankings, get_cosme_categories
        success, message, products = get_cosme_rankings(category, limit=limit)

        if success:
            return jsonify({
                'success': True,
                'data': {
                    'category': category,
                    'products': products,
                    'source': 'COSME 台灣',
                    'categories': get_cosme_categories()
                }
            })
        else:
            return jsonify({'success': False, 'error': message}), 500

    except Exception as e:
        logger.error(f"取得 COSME 排行榜失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/mybest_articles')
@login_required
def api_get_mybest_articles():
    """取得 mybest 推薦文章"""
    try:
        category = request.args.get('category', 'skincare')
        limit = request.args.get('limit', 10, type=int)

        if limit > 20:
            limit = 20

        from services.mybest_crawler import get_mybest_articles, get_mybest_categories
        success, message, articles = get_mybest_articles(category, limit=limit)

        if success:
            return jsonify({
                'success': True,
                'data': {
                    'category': category,
                    'articles': articles,
                    'source': 'mybest 台灣',
                    'categories': get_mybest_categories()
                }
            })
        else:
            return jsonify({'success': False, 'error': message}), 500

    except Exception as e:
        logger.error(f"取得 mybest 文章失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/mybest_latest')
@login_required
def api_get_mybest_latest():
    """取得 mybest 最新文章"""
    try:
        limit = request.args.get('limit', 20, type=int)

        if limit > 30:
            limit = 30

        from services.mybest_crawler import get_mybest_latest
        success, message, articles = get_mybest_latest(limit=limit)

        if success:
            return jsonify({
                'success': True,
                'data': {
                    'articles': articles,
                    'source': 'mybest 台灣'
                }
            })
        else:
            return jsonify({'success': False, 'error': message}), 500

    except Exception as e:
        logger.error(f"取得 mybest 最新文章失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


# ===== Web Search API =====

@ai_bp.route('/api/ai/web_search', methods=['POST'])
@login_required
def api_web_search():
    """
    AI 網路搜尋 - 使用 Ollama 進行智慧搜尋

    Request body:
    {
        "query": "搜尋關鍵字",
        "search_type": "general|news|shopping|trends",
        "num_results": 5,
        "model": "gemma3:4b" (optional)
    }
    """
    try:
        data = request.get_json()
        query = data.get('query', '')
        search_type = data.get('search_type', 'general')
        num_results = data.get('num_results', 5)
        model = data.get('model', None)
        save_to_history = data.get('save_to_history', True)

        if not query:
            return jsonify({'success': False, 'error': '請提供搜尋關鍵字'}), 400

        # 驗證搜尋類型
        valid_types = ['general', 'news', 'shopping', 'trends']
        if search_type not in valid_types:
            search_type = 'general'

        # 限制結果數量
        num_results = min(max(num_results, 1), 10)

        # 如果指定了模型，暫時切換
        original_model = ollama_service.model
        if model:
            ollama_service.model = model

        result = ollama_service.web_search(
            query=query,
            num_results=num_results,
            search_type=search_type
        )

        # 恢復原始模型
        if model:
            ollama_service.model = original_model

        if result.success:
            # 嘗試解析 JSON 結果
            response_data = {
                'raw_content': result.content,
                'model': result.model,
                'duration': round(result.total_duration, 2) if result.total_duration else None,
                'search_type': search_type,
                'query': query
            }

            try:
                content = result.content
                json_start = content.find('{')
                json_end = content.rfind('}') + 1
                if json_start >= 0 and json_end > json_start:
                    parsed = json.loads(content[json_start:json_end])
                    response_data['parsed'] = parsed
            except json.JSONDecodeError:
                pass  # 保持原始內容

            # 儲存到歷史記錄
            history_id = None
            if save_to_history:
                user_id = session.get('user_id')
                history_id = ai_history_service.save_generation(
                    generation_type='web_search',
                    output_content=result.content,
                    product_name=query,
                    input_keywords=[search_type],
                    input_style=search_type,
                    model_name=result.model,
                    generation_duration=result.total_duration,
                    created_by=user_id
                )
                response_data['history_id'] = history_id

            return jsonify({
                'success': True,
                'data': response_data
            })
        else:
            return jsonify({'success': False, 'error': result.error}), 500

    except Exception as e:
        logger.error(f"AI 網路搜尋失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/product_insights', methods=['POST'])
@login_required
def api_product_insights():
    """
    商品市場洞察分析

    Request body:
    {
        "product_name": "商品名稱",
        "include_competitors": true,
        "include_trends": true,
        "web_context": "網路搜尋結果（可選）",
        "model": "gemma3:4b" (optional)
    }
    """
    try:
        data = request.get_json()
        product_name = data.get('product_name', '')
        include_competitors = data.get('include_competitors', True)
        include_trends = data.get('include_trends', True)
        web_context = data.get('web_context', '')  # 網路搜尋結果
        model = data.get('model', None)
        save_to_history = data.get('save_to_history', True)

        if not product_name:
            return jsonify({'success': False, 'error': '請提供商品名稱'}), 400

        # 如果指定了模型，暫時切換
        original_model = ollama_service.model
        if model:
            ollama_service.model = model

        result = ollama_service.search_product_insights(
            product_name=product_name,
            include_competitors=include_competitors,
            include_trends=include_trends,
            web_context=web_context  # 傳入網路搜尋結果
        )

        # 恢復原始模型
        if model:
            ollama_service.model = original_model

        if result.success:
            response_data = {
                'raw_content': result.content,
                'model': result.model,
                'duration': round(result.total_duration, 2) if result.total_duration else None,
                'product_name': product_name
            }

            try:
                content = result.content
                json_start = content.find('{')
                json_end = content.rfind('}') + 1
                if json_start >= 0 and json_end > json_start:
                    parsed = json.loads(content[json_start:json_end])
                    response_data['insights'] = parsed
            except json.JSONDecodeError:
                pass

            # 儲存到歷史記錄
            history_id = None
            if save_to_history:
                user_id = session.get('user_id')
                history_id = ai_history_service.save_generation(
                    generation_type='product_insights',
                    output_content=result.content,
                    product_name=product_name,
                    input_keywords=['competitors' if include_competitors else '', 'trends' if include_trends else ''],
                    model_name=result.model,
                    generation_duration=result.total_duration,
                    created_by=user_id
                )
                response_data['history_id'] = history_id

            return jsonify({
                'success': True,
                'data': response_data
            })
        else:
            return jsonify({'success': False, 'error': result.error}), 500

    except Exception as e:
        logger.error(f"商品洞察分析失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/trend_keywords', methods=['POST'])
@login_required
def api_trend_keywords():
    """
    搜尋分類熱門關鍵字和趨勢

    Request body:
    {
        "category": "商品分類",
        "time_range": "day|week|month",
        "model": "gemma3:4b" (optional)
    }
    """
    try:
        data = request.get_json()
        category = data.get('category', '')
        time_range = data.get('time_range', 'week')
        model = data.get('model', None)
        save_to_history = data.get('save_to_history', True)

        if not category:
            return jsonify({'success': False, 'error': '請提供商品分類'}), 400

        # 驗證時間範圍
        if time_range not in ['day', 'week', 'month']:
            time_range = 'week'

        # 如果指定了模型，暫時切換
        original_model = ollama_service.model
        if model:
            ollama_service.model = model

        result = ollama_service.search_trend_keywords(
            category=category,
            time_range=time_range
        )

        # 恢復原始模型
        if model:
            ollama_service.model = original_model

        if result.success:
            response_data = {
                'raw_content': result.content,
                'model': result.model,
                'duration': round(result.total_duration, 2) if result.total_duration else None,
                'category': category,
                'time_range': time_range
            }

            try:
                content = result.content
                json_start = content.find('{')
                json_end = content.rfind('}') + 1
                if json_start >= 0 and json_end > json_start:
                    parsed = json.loads(content[json_start:json_end])
                    response_data['trends'] = parsed
            except json.JSONDecodeError:
                pass

            # 儲存到歷史記錄
            history_id = None
            if save_to_history:
                user_id = session.get('user_id')
                history_id = ai_history_service.save_generation(
                    generation_type='trend_keywords',
                    output_content=result.content,
                    product_name=category,
                    input_keywords=[time_range],
                    model_name=result.model,
                    generation_duration=result.total_duration,
                    created_by=user_id
                )
                response_data['history_id'] = history_id

            return jsonify({
                'success': True,
                'data': response_data
            })
        else:
            return jsonify({'success': False, 'error': result.error}), 500

    except Exception as e:
        logger.error(f"趨勢關鍵字搜尋失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/data_source_suggestion', methods=['POST'])
@login_required
def api_suggest_data_sources():
    """建議爬蟲資料來源"""
    try:
        data = request.get_json()
        product_name = data.get('product_name', '')
        product_category = data.get('product_category', '')
        model = data.get('model', None)

        if not product_name and not product_category:
            return jsonify({'success': False, 'error': '請提供商品名稱或分類'}), 400

        context = product_name if product_name else product_category

        system_prompt = """你是一位資深電商數據分析師，專精於台灣市場資料蒐集。
你熟悉各種可以爬取的公開資料來源，包括：
- 電商平台（momo、PChome、蝦皮、Yahoo購物等）
- 社群媒體（PTT、Dcard、Facebook 公開社團）
- 新聞媒體（Google News、各大新聞網站）
- 部落格與評測網站
- 價格比較網站
- 官方統計資料

請根據商品類型推薦適合的資料來源。"""

        prompt = f"""針對「{context}」這類商品/分類，請建議可以爬取的資料來源：

請提供 5-8 個建議，每個建議包含：
1. 網站名稱
2. 網站 URL（實際可訪問的網址）
3. 可獲取的資訊類型（價格/評價/趨勢/討論等）
4. 爬取難度（簡單/中等/困難）
5. 建議優先度（高/中/低）

請用以下格式回答：

【建議 1】
- 網站：XXX
- URL：https://...
- 資訊：XXX
- 難度：簡單
- 優先：高

請只回答實際存在且可訪問的台灣網站。"""

        # 如果指定了模型，暫時切換
        original_model = ollama_service.model
        if model:
            ollama_service.model = model

        result = ollama_service.generate(prompt, system_prompt=system_prompt, temperature=0.5)

        # 恢復原始模型
        if model:
            ollama_service.model = original_model

        if result.success:
            return jsonify({
                'success': True,
                'data': {
                    'suggestions': result.content,
                    'model': result.model,
                    'duration': round(result.total_duration, 2) if result.total_duration else None
                }
            })
        else:
            return jsonify({'success': False, 'error': result.error}), 500

    except Exception as e:
        logger.error(f"生成資料來源建議失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


# ══════════════════════════════════════════════════════════════
#  ICAIM — AI 競情中樞 (Intelligent Competitive Analysis & Intelligence Module)
#  2026-04-17 新增
# ══════════════════════════════════════════════════════════════

@ai_bp.route('/ai_intelligence')
@login_required
def ai_intelligence():
    """AI 競情中樞頁面"""
    return render_template('ai_intelligence.html', active_page='ai_intelligence')


@ai_bp.route('/api/ai/icaim/dashboard')
@login_required
def api_icaim_dashboard():
    """
    競情儀表板數據 API
    回傳：統計摘要 + 競品比價清單 + AI 決策日誌
    """
    try:
        from config import DATABASE_PATH
        from sqlalchemy import create_engine, text as sa_text

        engine = create_engine(DATABASE_PATH)

        # ── 統計摘要 ────────────────────────────────────────────
        # high_risk_count：MOMO 售價比 PChome 貴 > 15%（全量掃描）
        stats_sql = sa_text("""
            WITH latest_momo AS (
                SELECT p.i_code AS sku, pr.price AS momo_price,
                       ROW_NUMBER() OVER (PARTITION BY p.i_code ORDER BY pr.timestamp DESC) AS rn
                FROM products p
                JOIN price_records pr ON pr.product_id = p.id
                WHERE p.status = 'ACTIVE'
            ),
            high_risk AS (
                SELECT lm.sku
                FROM latest_momo lm
                JOIN competitor_prices cp
                    ON cp.sku = lm.sku
                   AND cp.source = 'pchome'
                   AND cp.expires_at > NOW()
                   AND cp.match_score >= 0.45
                WHERE lm.rn = 1
                  AND (lm.momo_price - cp.price) / cp.price * 100 > 15
            )
            SELECT
                (SELECT COUNT(*) FROM products WHERE status = 'ACTIVE')             AS total_skus,
                (SELECT COUNT(*) FROM competitor_prices
                  WHERE expires_at > NOW() AND source = 'pchome')                   AS valid_competitor_prices,
                (SELECT COUNT(*) FROM high_risk)                                    AS high_risk_count,
                (SELECT COUNT(*) FROM ai_price_recommendations)                     AS total_ai_recs,
                (SELECT MAX(crawled_at) FROM competitor_prices WHERE source='pchome') AS last_feeder_run
        """)

        # ── 競品比價（去重：每個 SKU 只取最新一筆 price_record）──
        competitor_sql = sa_text("""
            WITH latest_price AS (
                SELECT
                    p.i_code     AS sku,
                    p.name,
                    p.category,
                    pr.price     AS momo_price,
                    ROW_NUMBER() OVER (PARTITION BY p.i_code ORDER BY pr.timestamp DESC) AS rn
                FROM products p
                JOIN price_records pr ON pr.product_id = p.id
                WHERE p.status = 'ACTIVE'
            )
            SELECT
                lp.sku,
                lp.name,
                lp.category,
                lp.momo_price,
                cp.price           AS pchome_price,
                ROUND(((lp.momo_price - cp.price) / cp.price * 100)::numeric, 1) AS gap_pct,
                cp.match_score,
                cp.tags,
                cp.crawled_at
            FROM latest_price lp
            JOIN competitor_prices cp
                ON cp.sku = lp.sku
               AND cp.source = 'pchome'
               AND cp.expires_at > NOW()
            WHERE lp.rn = 1
            ORDER BY gap_pct DESC NULLS LAST
            LIMIT 200
        """)

        # ── AI 決策日誌 ─────────────────────────────────────────
        ai_sql = sa_text("""
            SELECT id, sku, name, strategy, confidence,
                   momo_price, pchome_price, gap_pct, sales_7d_delta,
                   reason, status, model_footprint, created_at
            FROM ai_price_recommendations
            ORDER BY created_at DESC
            LIMIT 50
        """)

        with engine.connect() as conn:
            stats_row  = conn.execute(stats_sql).fetchone()
            comp_rows  = conn.execute(competitor_sql).fetchall()
            ai_rows    = conn.execute(ai_sql).fetchall()

        # 格式化競品資料
        competitors = []
        for r in comp_rows:
            tags = r.tags or []
            if isinstance(tags, str):
                try:
                    tags = json.loads(tags)
                except Exception:
                    tags = []
            gap = float(r.gap_pct) if r.gap_pct is not None else 0.0
            competitors.append({
                'sku':          r.sku,
                'name':         r.name[:45] if r.name else '',
                'category':     r.category or '',
                'momo_price':   float(r.momo_price),
                'pchome_price': float(r.pchome_price),
                'gap_pct':      gap,
                'match_score':  float(r.match_score),
                'tags':         tags,
                'crawled_at':   r.crawled_at.strftime('%m/%d %H:%M') if r.crawled_at else '',
                'risk':         'HIGH' if gap > 15 else ('MED' if gap > 5 else 'LOW'),
            })

        # 格式化 AI 決策記錄
        ai_recs = []
        for r in ai_rows:
            footprint = r.model_footprint or {}
            analyst_fp   = footprint.get('analyst',    {}) if footprint else {}
            dispatch_fp  = footprint.get('dispatcher', {}) if footprint else {}
            ai_recs.append({
                'id':            r.id,
                'sku':           r.sku,
                'name':          r.name[:35] if r.name else '',
                'strategy':      r.strategy or 'promote',
                'confidence':    float(r.confidence) if r.confidence else 0.0,
                'momo_price':    float(r.momo_price)    if r.momo_price    else 0.0,
                'pchome_price':  float(r.pchome_price)  if r.pchome_price  else 0.0,
                'gap_pct':       float(r.gap_pct)        if r.gap_pct       else 0.0,
                'sales_7d_delta': float(r.sales_7d_delta) if r.sales_7d_delta else 0.0,
                'reason':        r.reason or '',
                'status':        r.status,
                # AI 模型足跡（供儀表板顯示推理開銷）
                'analyst':       analyst_fp.get('model', 'hermes3'),
                'hermes_duration': analyst_fp.get('duration_sec', 0),
                'hermes_tokens': analyst_fp.get('tokens', 0),
                'dispatcher':    dispatch_fp.get('model', 'NIM'),
                'nim_tokens':    dispatch_fp.get('total_tokens', 0),
                'created_at':    r.created_at.strftime('%m/%d %H:%M') if r.created_at else '',
            })

        last_feeder = (
            stats_row.last_feeder_run.strftime('%Y-%m-%d %H:%M')
            if stats_row.last_feeder_run else '尚未執行'
        )

        return jsonify({
            'success': True,
            'stats': {
                'total_skus':              int(stats_row.total_skus or 0),
                'valid_competitor_prices': int(stats_row.valid_competitor_prices or 0),
                'high_risk_count':         int(stats_row.high_risk_count or 0),
                'total_ai_recs':           int(stats_row.total_ai_recs or 0),
                'last_feeder_run':         last_feeder,
            },
            'competitors': competitors,
            'ai_recs':     ai_recs,
        })

    except Exception as e:
        logger.error(f"[ICAIM] dashboard API 失敗: {e}")
        return jsonify({'success': False, 'error': str(e)}), 500


@ai_bp.route('/api/ai/icaim/trigger', methods=['POST'])
@login_required
def api_icaim_trigger():
    """
    手動觸發一次完整 ICAIM 分析（非同步，背景執行）
    立即回傳 202，後台跑 Hermes → NemoTron → Telegram
    """
    import threading

    def _run_pipeline():
        try:
            from config import DATABASE_PATH
            from sqlalchemy import create_engine
            from services.hermes_analyst_service import HermesAnalystService
            from services.nemoton_dispatcher_service import NemotronDispatcher as NemotronDispatcherService
            from services.notification_manager import NotificationManager

            engine = create_engine(DATABASE_PATH)
            hermes_svc = HermesAnalystService(engine=engine)

            t0 = datetime.now()
            result = hermes_svc.run()
            hermes_duration = round((datetime.now() - t0).total_seconds(), 1)

            logger.info(f"[ICAIM][手動觸發] Hermes 完成 threats={len(result.threats)} 耗時={hermes_duration}s")

            if result.threats:
                hermes_stats = {
                    'model':        'hermes3:latest',
                    'duration_sec': hermes_duration,
                    'tokens':       result.hermes_tokens,
                }
                notifier = NotificationManager()
                dispatcher = NemotronDispatcherService(
                    notification_manager=notifier, engine=engine
                )
                dispatch_result = dispatcher.dispatch(result.threats, hermes_stats=hermes_stats)
                logger.info(f"[ICAIM][手動觸發] NIM dispatch 完成: {dispatch_result}")
            else:
                logger.info("[ICAIM][手動觸發] 無威脅商品，跳過 NIM dispatch")

        except Exception as ex:
            logger.error(f"[ICAIM][手動觸發] 背景執行失敗: {ex}")

    thread = threading.Thread(target=_run_pipeline, daemon=True)
    thread.start()

    return jsonify({
        'success': True,
        'message': '分析已在背景啟動，約 30~60 秒後完成，請重新整理儀表板查看結果',
    }), 202